Cerebras Systems, a Silicon Valley-based startup that produces artificial intelligence (AI) chips, has released open source models similar to ChatGPT for free use by researchers and businesses.
Cerebras created seven models, all trained on its AI supercomputer called Andromeda. These models range from smaller 111 million parameter language models to larger 13 billion parameter models. By releasing these models, Cerebras hopes to encourage greater collaboration within the research and business community.
Cerebras’ CEO, Andrew Feldman, commented on the importance of maintaining openness within the AI community. He noted that while there is currently a trend towards closing off AI resources, the progress that has been made within the community has been largely due to the open sharing of resources.
Cerebras’ models with more parameters are capable of performing more complex generative functions. OpenAI’s chatbot, ChatGPT, launched in late 2022 with 175 billion parameters and has since generated significant interest and funding within the field of AI.
Cerebras’ models are capable of running on a variety of devices, from smartphones and smart speakers to PCs and servers. However, complex tasks such as large passage summarization require the use of larger models.
While there is a belief that bigger models are better, chip consultant Karl Freund points out that smaller models can be accurate with more training. Despite this, Cerebras’ AI training architecture has allowed for the creation of larger models in a relatively short period of time.
Though Nvidia Corporation currently dominates the AI chip market, startups like Cerebras are beginning to take market share. Additionally, the models created by Cerebras can also be used on Nvidia systems for further training or customization, Reuters news report said.